PR #3837 llama : refactor graph build code

llama : factor out ggml-alloc from graph graph build functions

ggerganov committed 2 years ago

metal : disable kernel load log

ggerganov committed 2 years ago

llama : factor out tensor offloading outside the build call (wip)

ggerganov committed 2 years ago

llama : offload rest of the models

ggerganov committed 2 years ago

llama : update offload log messages to print node index

ggerganov committed 2 years ago

llama : comments

ggerganov committed 2 years ago

llama : support offloading result_norm + comments

ggerganov committed 2 years ago

llama : factor graph input into a function

ggerganov committed 2 years ago

llama : do tensor offload only with CUDA

ggerganov committed 2 years ago

llama : fix res_norm offloading

ggerganov committed 2 years ago

llama : try to optimize offloading code

ggerganov committed 2 years ago

llama : fix non-CUDA build

ggerganov committed 2 years ago

llama : try to fix build

ggerganov committed 2 years ago

llama : move refact in correct place + optimize graph input

ggerganov committed 2 years ago

llama : refactor tensor offloading as callback

ggerganov committed 2 years ago

llama : add layer index to all tensor names

ggerganov committed 2 years ago

llama : add functional header

ggerganov committed 2 years ago

llama : comment

ggerganov committed 2 years ago

llama : remove obsolete map for layer counting

ggerganov committed 2 years ago

llama : add llm_build helper functions (#3848)

ggerganov committed 2 years ago

Merge branch 'master' into llama-refactor

ggerganov committed 2 years ago

llama.cpp llama : refactor graph build code #3837 Merged